ftynse wrote: Since this is auto-generated, how can we evaluate the quality of the highlightling? Maybe you can post some before/after pictures for a dozen of examples drawn from tests? https://github.com/llvm/llvm-project/pull/166406